NYU Reactive Gripper : An Implementation

We consider the problem of grasping an unknown polygonal at object using a parallel jaw gripper. Our design equips a standard gripper with several light-beam sensors (close to each jaw) and employs a control scheme based on a reactive grasping algorithm. This is done by probing the object to locate a good grasp position, and then grasping, without moving the object signiicantly. The goal is to do as little motion as possible to nd a grasp. In this paper, we discuss an implementation of this device using NYU's MOSAIC robot, following a quick overview of the underlying reactivity principle. 1 What is a Reactive Algorithm? By a reactive algorithm, we mean an algorithmic scheme where a robot's sensors determine directly the actions of the actuators. However, note that the actuators themselves may interact with the sensors (e.g., by moving them or occluding them, etc.) to close a feedback loop and thus causing further goal-driven as well as corrective actions. In the simplest picture, a set of sensors determine a binary vector, where each bit of the vector may represent whether a sensor detects a signal or not (\on/oo sensor"), whether the value detected by one signal is higher than another or a predetermined threshold, etc. Each possible value of the binary vector determines the movement of a single actuator immediately. The necessary actions for a given value of the vector are expressed in terms of a table with one entry per vector value. In a more complicated picture, we allow nitely many states with a table per state. In the table, in addition to the actuator moves, we also have an entry for state transitions. Ideas similar to ours have also been studied by cyberneticists who built simple, reeex-based artiicial animals to demonstrate the possibility of mechanical behavior and learning. For example Grey Walter's light-sensitive turtles and Braitenberg's \vehicles of desire." Another eeort of similar nature is Rod Brook's 2] subsumption architecture and the insect species (e.g., Atilla the ant). Unlike these approaches, we view our problems as engineering problems and solve them without any inspiration or analogy from an anthropomorphic solution. Our endeavors are much closer to the ones in control theory (discrete event system